HIGHLIGHTED ARTICLE | INVESTIGATION

Epistatic Networks Differentially Modulate Arabidopsis Growth and Defense

Baohua Li,*,†,1 Michelle Tang,*,‡,1 Céline Caseys,*,1 Ayla Nelson,* Marium Zhou,* Xue Zhou,* Siobhan M. Brady,‡ and Daniel J. Kliebenstein*,§,2 *Department of Plant Sciences, University of California, Davis, California 95616, †College of Horticulture, Northwest A&F University, Yangling, Shaanxi, 712100, China, ‡Department of Plant Biology and Center, University of California, Davis, California 95616, and §DynaMo Center of Excellence, University of Copenhagen, DK-1871 Frederiksberg C, Denmark ORCID IDs: 0000-0001-7235-0470 (B.L.); 0000-0002-8779-8697 (M.T.); 0000-0003-4187-9018 (C.C.); 0000-0002-2586-0264 (X.Z.); 0000-0001-9424-8055 (S.M.B.); 0000-0001-5759-3175 (D.J.K.)

ABSTRACT Plants integrate internal and external signals to finely coordinate growth and defense for maximal fitness within a complex environment. A common model suggests that growth and defense show a trade-offs relationship driven by energy costs. However, recent studies suggest that the coordination of growth and defense likely involves more conditional and intricate connections than implied by the trade-off model. To explore how a transcription factor (TF) network may coordinate growth and defense, we used a high-throughput phenotyping approach to measure growth and flowering in a set of single and pairwise mutants previously linked to the aliphatic glucosinolate (GLS) defense pathway. Supporting a link between growth and defense, 17 of the 20 tested defense- associated TFs significantly influenced plant growth and/or flowering time. The TFs’ effects were conditional upon the environment and age of the plant, and more critically varied across the growth and defense for a given . In support of the coordination model of growth and defense, the TF mutant’s effects on short-chain aliphatic GLS and growth did not display a simple correlation. We propose that large TF networks integrate internal and external signals and separately modulate growth and the accumulation of the defensive aliphatic GLS.

KEYWORDS ; glucosinolates; plant defense; plant growth; transcription factor

ROWTH and defense are essential biological processes comes from the observations that some constitutive defense Gnecessary for plant survival. Optimizing fitness requires mutants generally grow smaller and suffer from yield and/or plants to coordinate growth and defense phenotypes in re- fitness losses. sponse to specific environments. Efforts to understand the A developing model is emerging from research in ecology, relationship between plant growth and defense are often evolutional biology, and molecular that suggests a modeled as a trade-off where resistance and growth are a dynamic relationship between plant defense and growth cost on each other (Karban and Baldwin 1997; Agrawal (Singh et al. 2002; de Lucas et al. 2016). This model began 2011a,b; Huot et al. 2014). This model assumes that the with multiple reports showing that defense metabolism does available resources for plants are limited, suggesting that not show a universal negative correlation with plant growth any flux of resources, energy and elements into plant defense (Bergelson and Purrington 1996; Almeida-Cortez et al. 1999; would be at the cost of plant growth. Support for this model Koricheva 2002). Additionally, mechanistic manipulations of defense metabolism can show little to no effect on plant growth. Further, diminutive constitutive defense mutants can Copyright © 2020 by the Genetics Society of America doi: https://doi.org/10.1534/genetics.119.302996 have their growth rescued by second-site that Manuscript received August 7, 2019; accepted for publication December 17, 2019; maintain the constitutive defense (Hemm et al. 2003; Paul- published Early Online December 18, 2019. Supplemental material available at figshare: https://doi.org/10.25386/genetics. Victor et al. 2010; Züst et al. 2011; Joseph et al. 2013; 11388195. Campos et al. 2016; Kliebenstein 2016). Finally, support for 1These authors contributed equally to this study. fi fi 2Corresponding author: University of California, Davis, One Shields Ave., Davis, CA this model came from the identi cation of speci c transcrip- 95616. E-mail: [email protected] tion factors (TFs) that simultaneously increase growth and

Genetics, Vol. 214, 529–541 February 2020 529 defenses (Campos et al. 2016; Wang et al. 2018). Together, the majority of their accumulation is a steady-state response to the these observations suggest that the relationship between environment with between natural accessions plant defense and growth is a complex and active internal having larger effects on accumulation than specificsignalpercep- decision process involving regulatory and signaling pathways tion events (Kliebenstein et al. 2002b; Mikkelsen et al. 2003; in planta (Kliebenstein 2016; Züst and Agrawal 2017). Wentzell and Kliebenstein 2008; Chan et al. 2011; Guo et al. Several recent studies provide important insights about the 2013; Pangesti et al. 2016). Thus, we focused on using genetic roles TFs play in integrating and transducing internal and perturbation of aliphatic short-chain (SC) GLS accumulation, the external signals essential to growth and defense (Singh et al. major form of aliphatic GLS, as an optimal way to test the sys- 2002; Pajerowska-Mukhtar et al. 2012; Lozano-Duran et al. tem and how epistasis may link defense, growth, and flowering. 2013; Fan et al. 2014; de Lucas et al. 2016). Mutations in To test the conditional model of growth and defense, we used JAZs (transcriptional repressors injasmonicacidsignalingpath- high-throughput phenotyping to measure growth and flowering way) in combination with altered photoreceptor PhyB result in time of 20 single TF mutants and 48 double mutants previously fast-growing plants with enhanced plant defense responses shown to influence the Arabidopsis aliphatic GLS pathway in (Campos et al. 2016). In rice, the Ideal Plant Architecture 1 TF two environments (Sønderby et al. 2007; Li et al. 2018). We activates yield-related while promoting both plant de- show that 17 of the 20 TFs significantly influence plant growth fense and rice yield (Jiao et al. 2010; Wang et al. 2018). These and/or plant flowering time. While most TFs influence growth case studies demonstrate that growth and defense are under and defense, there was no clear correlation between the ali- complex regulation and that TFs can be key integrators to co- phatic SC GLS accumulation and plant growth and flowering ordinate these two important biological processes. However, it time in our study. This indicates that each TF has specificand is not clear if these examples are isolated instances or a more independent influences on both growth and aliphatic GLS. Our generalizable view of how TFs modulate growth and defense. findings support the coordination model for the relationship Systemic studies on TFs and their regulatory networks are between plant growth and aliphatic GLS, and provide novel needed to test how TFs may or may not coordinate between insights on how these critical and complex biological processes plant defense and growth in diverse environmental settings. are integrated to optimize fitness in different environments. To explore the interplay between plant growth and de- fense, we used the well-characterized plant secondary met- abolic pathway, the Arabidopsis methionine-derived aliphatic Materials and Methods glucosinolate (GLS) (Sønderby et al. 2010b). Aliphatic GLS Plant materials, growth conditions, and provide defense against numerous herbivorous in experimental design both the laboratory and the field (Lambrix et al. 2001; Schlaeppi et al. 2008; Kos et al. 2012; Beran et al. 2014; The Arabidopsis thaliana transfer DNA (T-DNA) insertion Falk et al. 2014; Kerwin et al. 2015, 2017; Zalucki et al. lines of the 20 TFs were ordered from Arabidopsis Biological 2017). Aliphatic GLS are critical for fitness in Brassicales, Resource Center (Sussman et al. 2000; Alonso et al. 2003) and their specific composition and accumulation across de- and homozygous lines were validated in previous studies velopmental stages are intricately controlled by genetic var- (Sønderby et al. 2010a; Li et al. 2014). The full description iation and pleiotropically influence a variety of growth- and of how the 48 double mutants were generated and validated defense-related pathways (Kliebenstein et al. 2002a; Bidart- and planted is provided in a previous study (Li et al. 2018). Bouzat and Kliebenstein 2008; Wentzell and Kliebenstein Briefly, the Arabidopsis plants were grown in two indepen- 2008; Burow et al. 2010; Kerwin et al. 2011; Züst et al. dent chambers with 16-hr light at 100–120 mEi light intensity 2011). Since aliphatic GLS contain both sulfur and nitrogen, with temperature set at a continuous 22°. The two growth flux-based modeling suggests that they are expensive to pro- chambers were set to identical abiotic environments but con- duce as they can accumulate at high concentrations. How- tain dramatically different biotic environments: one pest- ever, mutants missing these compounds have at most a slight free, with a clean Controlled Environment Facility (CEF) change in early growth, indicating that aliphatic GLS do not chamber as a typical laboratory growth condition; and one display a tight trade-off with growth (Paul-Victor et al. 2010; with an endogenous pest population, with a stress Life Sci- Züst et al. 2011; Bekaert et al. 2012). Instead, the regulatory ences Addition building (LSA) chamber mimicking more nat- complexity of the aliphatic GLS better fits with the coordina- ural growth condition. Seeds were imbibed in water at 4° for tion model. For example, a large-scale yeast one-hybrid study 3 days and sown into Sunshine Mix (Sun Gro Horticulture). identified a large collection of TFs that bind the aliphatic GLS Seedlings were thinned to one plant per pot (6 cm 3 5 cm) at promoters and regulate aliphatic GLS accumulation, 7 days after planting. The pots were in 36 cell flats and each directly arguing against a single dominant TF model (Li et al. flat was considered as a blocking unit. All four 2014). Further, these TFs showed extensive epistatic interac- (, two single mutants, and one double mutant) for tions in influencing the accumulation of GLS (Li et al. 2018). a double mutant combination were grown together in a sin- This suggests that the regulation of this defense pathway is gle flat along with genotypes for other double mutant com- highly complex. While aliphatic GLS levels are influenced by binations, such that there was only a single representative of perception of jasmonic acid and other herbivory-related signals, each genotype within the flat. This was replicated across

530 B. Li et al. eight flats with randomization within each flat to create eight To determine the power to identify significant epistatic and replicates of each genotype for each comparison. This exper- epistatic 3 conditional terms given our experimental design, iment was conducted independently in the clean CEF and we used the pwr.f2.test function from the pwr package stress LSA chamber to generate a minimum of 16 biological (https://github.com/heliosdrm/pwr). Using the median num- repeats in total per genotype. The total combination of all ber of error degrees of freedom across all models of 542 and a genotypes required 24 flats per growth chamber. All genotypes significance level of 0.05, we have a 0.8 power to detect sig- were independently randomized within a flat. nificant epistatic and epistatic * conditional terms if the terms Flowering time and plant growth measurement effect size (f2) is 0.0145 or higher. This is within the range of effect sizes linked to the epistatic or epistatic 3 conditional All the plants were checked daily, and the flowering time, the first terms in our models (Supplemental Material, Figure S1). This fl day of owering after planting, was recorded for each of the plants. indicates that we had sufficient power to identify epistatic 3 Pictures of all the plants were taken every other day from day 9 to conditional using this experimental design (Figure S1). fl 27, whensomeof theearly owering lines in the stress LSAchamber In these analyses, we considered the 20 models for main fl started to ower.The circling area of the rosette of each plant in each effects or 48 models for the different epistatic comparisons as growth condition was manually measured in ImageJ across the our independent tests. Using a nominal P value across 48 tests days. The total circling leaf area was used as an indicator of the plant would suggest between two and three potential false positives fl size. The plant owering time and plant sizes were normalized per trait for the epistatic analyses. However, there was on av- relative to Col-0, and further visualized using the iheatmapr erage 20 (62 SE) tests below 0.05 for the main epistatic inter- package in R software (R Development Core Team 2014). actions per trait and on average 8 (61SE)testsbelow0.05for Statistics the chamber 3 epistasis terms. This is much higher than the two or three expected false positives. To maximize our ability to Each pair of TFs were separately analyzed using a repeated compare across traits and genes, we have opted to utilize the measures general linear model with lme4 to test for epistasis less conservative nominal P values for the comparisons. Fur- affecting plant growth (Bates et al. 2015). For this analysis each ther, all P values and least-square means are reported to allow double mutant was tested using both single mutants and wild for direct assessment of the results (Tables S1 and S2). type grown concurrently with the following model: yabc = m +Aa +Bb +Chc +Dn +Aa3Bb +Aa3Chc +Bb3Chc +Aa3Dn + Calculation of epistasis value 3 3 3 3 3 3 3 3 Bb Dn+Chc Dn +Aa Bb Dn +Aa Chc Dn +Bb Chc Dn + To study the effect of epistasis, we use epistasis value to 3 3 3 e Aa Bb Chc Dn +Pi + abc,wherePi istherandomvariableof describe the direction and strength of the epistasis. The Plant, the experimental unit from which measurements were epistasis value was calculated by normalizing the difference taken over time. e , the error term is assumed to be normally abc of observed double mutant vs. the predicted dou- 2 distributedwithmean0andvariancese. In this model, y abc ble mutant phenotype, assuming additivity of the single mu- denotes the rosette area of each plant, genotype A represents tants, then normalized to the wild type as reported before the presence or absence of a T-DNA insert in one TF (wild (Segrè et al. 2005; Li et al. 2018). The phenotype for wild type vs. mutant of A), and genotype B represents the pres- type was set as w, mutant TFa as a, mutant TFb as b, and ence or absence of a T-DNA insert in another TF gene (wild type double mutant TFa/TFb as ab. The epistasis value is calcu- vs. mutant of locus B) in the double mutant from chamber Ch c lated as {ab – [w + (a2w) + (b2w)]/w}. If the epistasis (clean CEF chamber or stress LSA chamber). Time is represented value is positive, this shows evidence for synergistic epistasis, by days, D . This model was run individually for each double n while antagonistic epistasis is reflected in negative values. mutant test and only the single mutants and wild-type Col- Larger the epistasis value signifies stronger the epistasis ef- 0 grown alongside the double mutants in the same flats were fects. The epistasis value were further visualized using the used in each model. This means that each double mutant test iheatmapr package in R software (R Development Core Team uses matched wild-type and single mutant control plants. Simi- 2014). larly, a general linear model was used to test epistasis affecting

flowering time in each pair of TFs: yabc = m +Aa +Bb +Chc + Data availability A 3B +A3Ch +B3Ch +A3B 3Ch +Ch:U + e , a b a c b c a b c c i abc All genotypes are available upon request. File S1 contains where U is the random effect of blocks nested in chamber Ch . i c all supplemental figures and tables mentioned within the text. The ANOVA table, least-square means, and SE for each genotype Supplemental material available at figshare: https://doi.org/ by treatment combinations were obtained with lme4 and 10.25386/genetics.11388195. emmeans packages in R (Searle et al. 1980; Bates et al. 2015). The type III sums of squares from these models were used to calculate the variance and percent variance attributable to each Results term in the model. For the percent variance, this was calculated Conditional growth effects of the 20 TF mutations by comparing to the total variance in the model as the denom- inator. All network representations were generated using Cyto- The 20 selected TFs were originally identified as binding scape version 2.8.3 (Shannon et al. 2003). aliphatic GLS-related promoters and their mutants influence

Growth and Defense Coordination by TFs 531 Figure 2 The effects of the 20 TFs on plant growth. The heatmap dis- plays the fold change of plant size in single mutants, from day 9 to 27, in both the clean CEF chamber and stress LSA chamber. Red shows in- creased plant size and blue decreased plant size in comparison to Col-0. The columns on the right display the statistical significance (purple, significant P , 0.05; gray, not significant) for each term in the statistical model as listed at the bottom using the repeated measure model. Spe- cific framed genotypes have mean phenotypes shown for reference in Figure S2.

with a blend of biotic pressures. Further, we acknowledge potential differences in the abiotic environments due to var- iation in ventilation or other parameters that were not con- trollable. Additionally, the mutant collection provides a large range of perturbation in defense independently of the different chambers that is similar to or even larger than the range of GLS variation in natural accessions (Figure 1) (Kliebenstein et al. 2001; Chan et al. 2010). This mutant Figure 1 Extensive range of variation in short-chain GLS content of wild collection generates a wider range of SC GLS defense varia- type, single mutants, and double mutants in CEF and LSA. The average fi absolute SC GLS content across all the genotypes in each growth cham- tion than observed in multiple years of eld trials for Col-0 ber is shown. Wild type (Col-0) is highlighted as a green dashed line, (Kerwin et al. 2015). Thus, the LSA and CEF chambers test while all the other genotypes are shown as gray solid lines. how the growth and defense in the mutant collection may or may not be sensitive to complex environmental perturbations aliphatic GLS accumulation (Li et al. 2014, 2018). We focused (Figure 1). on the aliphatic GLS as they represent 90% of the total GLS Growth was measured every other day from 9 days post- content within Arabidopsis leaves and therefore the majority germination to 27 days, at which time the leaves were over- of the metabolic flux. To measure the plant size in this mutant lapping. Plants were organized in a randomized complete collection, we utilized digital image analysis on plants from block design with eight measurements per genotype per two chambers, coded by their location in the CEF (clean chamber (Li et al. 2018). The plant growth measurement chamber) and LSA (stress chamber) (Li et al. 2018). The spans the majority of vegetative growth providing a dynamic two chambers, both built by Conviron, are similar in size, analysis of growth. Combining the data with repeated with identical light and humidity settings. The two chambers measure linear models, we tested for significant effects of differ in their pest population: the clean CEF chamber is ster- all single gene TF mutants on growth across the conditions ilized monthly after each planting while the stress LSA cham- (Figure 2, Figure S2, and Supplemental Data Sets 1 and 2). ber contains longer-lived plants that maintain endogenous The models indicated that the influence of these TFs on populations of pests likes gnats, various flea beetles, growth is highly conditional on both plant age and growth aphids, etc. This creates a chamber that presents the plant chamber (Figure 2 and Figure S3). Of the 20 aliphatic GLS TF

532 B. Li et al. Figure 4 Epistatic effects on plant growth. Heatmap of epistasis values for all pairwise TF combinations from day 9 to 27 in each treatment condition. The genotypes are clustered using hierarchical clustering. The columns to the right of the heatmap show epistatic terms as significant (purple) or not significant (gray) (P , 0.05). Framed genotypes are plotted in Figure 5.

mutants, 16 had a significant influence on growth. Strong positive growth effects were observed more often in the stress LSA chamber than in the clean CEF chamber for a number of TFs, like HMGBD15, AT5G52020, ZFP4, CBF4, NAC102, and GBF2 (Figure 2). In contrast, ABF4 had strong negative Figure 3 Epistatic between TF genes modeled for plant growth effects in both chambers. In addition to differences growth. Epistatic networks for (A) TF 3 TF 3 chamber, (B) TF 3 TF 3 3 3 3 between the growth chambers, most of the TFs had differen- day, and (C) TF TF chamber day. A line connecting two TFs fl correspond to a significant epistatic interaction. Node color indicates in- tial in uence across plant age. Some mutants combined these dividual TFs’ significance. Sky blue means that the individual TF has a conditionalities as illustrated by ilr3, which had a transition significant genotype 3 chamber 3 day term, green means significant at 21 days postgermination from positive to negative growth in both genotype 3 chamber 3 day and genotype 3 chamber terms, effects, but only in the LSA chambers (Figure S3). This con- fi 3 3 3 while red means signi cant in genotype chamber day, genotype ditionality illustrates the importance of large-scale phenotyp- chamber, and genotype terms. Gray means not significant. (A–C) To maximize comparison across publications, the TFs are laid out in the ing across different conditions to generate a broad view of network as groups A–G based on their modular effect on aliphatic GLS how mutational effects may change dynamically (Figures S2 accumulation (Li et al. 2018). and S3). In an energetic trade-off, we would expect that most

Growth and Defense Coordination by TFs 533 Figure 6 TF effects on flowering time. The heatmap displays the fold change of flowering time in the single mutants in both clean CEF cham- ber and stress LSA chamber. Red shows increased flowering time and blue decreased flowering time in comparison to Col-0. The columns on the right display the statistical significance (purple, significant P , 0.05; gray, not significant) for each term in the statistical model as listed at the bottom.

of these TF mutations should lead to smaller plants as these mutations increase aliphatic GLS content (Li et al. 2014). In contrast, the mutants had a mix of positive and negative growth effects. In addition, three mutants with strong ali- phatic GLS phenotypes, ant, myb28, and myb29, have little to no detectable main effect on growth. One proposed alter- native explanation for the myb28 and myb29 mutant is that these mutations often have a twofold increase in indolic GLS accumulation that may balance the loss in aliphatic GLS. Figure 5 Epistatic growth phenotypes at day 15. The average ro- However, because aliphatic GLS represent 90% of the total sette size on day 15 for three pairs of TFs in the clean (CEF) and fl stress (LSA) chambers. Each TF’s pairwise combination has a wild- GLS production, the GLS ux is still reduced by 80% in these type (Col-0) genotype. Letters indicate genotypes with significantly mutants. The links between plant growth and defense via the different plant sizes. (P , 0.05 using post hoc Tukey’s test). SE was aliphatic GLS in this TF collection supports the coordination calculated for eight samples per genotype and treatment. Day model of growth and defense. 15 was chosen to provide a common date across which to illustrate key differences. (A) Rosette size of single and double mutants of Dynamic epistatic networks underlying plant growth rap2.6l and erf9. (B) Rosette size single and double mutants of ilr3 and hmgbd15. (C) Rosette size single and double mutants of myb28 We previously generated a set of 48 pairwise mutant combi- and myb29. nations from 20 TFs controlling aliphatic GLS that form an

534 B. Li et al. Figure 7 Epistatic network and flowering time effects. (A) Heatmap of epistasis values for all pairwise combinations individually in both treatment conditions. The genotypes are clustered using hierarchical clustering. The columns to the right display interactions as significant (purple) or not significant (gray) (P , 0.05). To maximize comparison across publications, the TFs are laid out in the network as groups A–G based on their modular effect on aliphatic GLS accumulation (Li et al. 2018). (B) A representation of the significant epistatic networks for flowering time. Solid lines indicate significant TF 3 TF interaction. Dotted lines indicate significant treatment 3 TF 3 TF interaction. Node color indicates individual TFs significance: sky blue, TF; green, treatment 3 TF; yellow, TF + treatment 3 TF; gray, not significant. (C) Visualization of individual epistatic variance components within the genetic network for flowering time. The width of the line connecting two TFs is proportional to the variance linked with the TF 3 TF term for that specific interaction.

Growth and Defense Coordination by TFs 535 extensive antagonistic epistatic network (Sønderby et al. 2007; Li et al. 2018). As these TFs affect growth individually, we used repeated measures linear models to assess if the epistatic effects were as prevalent on growth as on aliphatic GLS accumulation. We plotted the potential epistatic effects for each mutant pair and represented these interactions as a connectivity plot based on the mutants’ main effects on ali- phatic GLS (Figure 3, Figure 4, and Supplemental Data Set 3 and, 4). This analysis identified extensive epistatic interac- tions influencing growth. As observed in the single mutants, the epistatic interactions were equally conditional across on- togeny and growth condition (Figure 3 and Figure 4). These epistatic interactions followed a similar clustering as that observed when using GSL phenotypes for clustering. Partic- ularly, the TFs in group F, ERF9, DF1, ABF4, and Rap2.6L, showed frequent epistatic interactions with each other for growth. Thus, this set of TFs identifies and illustrates an ep- istatic system that influences plant growth depending on age and growth condition. The effects of these TFs on growth are more conditional than the aliphatic GLS, suggesting that while the traits of growth and defense are both controlled by the epistatic network, they are independent regulatory outputs. Conditional epistatic networks underlie plant growth To quantify the epistatic effects on plant growth, we used a previously established epistasis value to measure the direction and magnitude of each epistatic interaction (Li et al. 2018). Briefly, we subtracted the measured double mutant phenotype from the predicted double mutant phenotype under an addi- tive model. This value was then normalized to the wild-type phenotype. This epistasis value was measured for each pair of mutants for all the growth data (Supplemental Data Set 5). The epistasis value will be positive when there is synergistic epistasis and negative for antagonistic epistasis (Figure 4). Previous work showed that epistasis for SC GLS, the dominant form of aliphatic GLS, was almost entirely negative/antagonis- tic (Li et al. 2018). Unlike SC GLS, epistasis for growth was a mix of antagonistic and synergistic values that shift depending upon the conditions and developmental stages. For example, rap2.6l was involved in several epistatic interactions with other TFs from group F that were positive in the LSA growth condition but negative in the CEF growth condition (i.e., rap2.6/erf9). In contrast, epistatic interactions involving hmgbd15 had negative interactions in the LSA and positive Figure 8 Lack of relationship between mutant effects on defense and interactions in the CEF growth conditions (i.e., ilr3/hmgbd15) growth. The main additive effect of the TF mutants in stress LSA chamber (Figure 4 and Figure 5). This argues that this epistatic network were selected to illustrate the absence of a relationship between growth, of TFs influences growth in both conditions, but that the envi- fl owering time, and SC GLS. The predicted linear trend line and the ronmental signals in the two conditions permeate differently statistical test results are shown. The axis shows the fold effect of the average mutant phenotype in comparison to the respective wild-type through the TF network to generate variable growth outputs. value. (A) Comparison of average gene effects on SC GLS and flowering fl fl time (r = 20.006, P = 0.980). This result was nonsignificant in both the Epistatic networks in uence owering independent of presence and absence of the outlier point for SC GLS. (B) Comparison of tested environments average gene effects on SC GLS and growth (r = 0.131, P = 0.583). This fl result was nonsignificant in both the presence and absence of the outlier To test if these TFs and their network in uence reproduction, point for SC GLS. (C) Comparison of average gene effects on flowering we measured flowering time in all of the plants from all of time and growth (r = 20.867, P , 0.001). the genotypes in the two contrasting chambers, and showed

536 B. Li et al. that 12 of the 20 tested TFs significantly influenced flowering (Figure 6). However, unlike growth, the effects on flowering time were almost entirely toward early flowering in both chambers, with only ANT, ILR3,andERF9 having an environ- mental conditionality (Figure 6 and Figure S4). Plotting the epistatic effects for flowering time showed predominantly syn- ergistic epistasis with 18 statistically significant interactions and only two interactions having environmental conditionality (Figure 7, A and B). This indicates that the genetic control of flowering time is less influenced by the environmental condi- tions than are either growth or aliphatic GLS. To further visu- alize how the epistatic variance was influenced by the network topology, we mapped the epistatic variance for the plant flow- ering (Figure 7C). Using previously ascribed groupings of the TFs based on their GLS phenotypes, TFs in groups A and B have more significant epistatic interactions partly overlapping with the epistatic interactions of growth phenotype, and TFs with strong main effects on flowering time also have higher genetic variance in their epistatic interactions. Connections between defense and growth To further explore the connection between plant defense and development, we systemically tested for associations between growth and defense in this data. These TFs predominantly influence the accumulation of SC GLS, and we used this as our quantification for defense (Supplemental Data Set 1) (Li et al. 2018). The mutant effects of the 20 TFs on SC GLS, plant growth and flowering time in both chambers were calculated by taking the phenotypic difference between mutant and wild type and normalized by the wild-type value. We then tested for a relationship between growth and defense by test- ing for correlations between the mutant effects using these traits (Figure 8 and Supplemental Data Set 6). In contrast to the expectation that an energetic trade-off model was solely driving this system, the TFs’ effects on the accumulation of SC GLS and growth/flowering were largely uncorrelated (Figure 8, A and B). This was equally true in both the stress and clean chamber, indicating that stress did not illuminate a hidden relationship. As expected, there was a negative correlation between growth and flowering. Our findings show a highly complex relationship between plant defense and plant growth. The connection of single mutant and double mutant effects Figure 9 Relationship of the main and epistatic effects for TF genes. The average epistatic effect of a TF across all its pairs in the stress LSA cham- Given the low correlation between the SC GLS accumulation ber was calculated and compared to the TFs main effects for SC GLS, and plant growth, we proceeded to investigate if there may be fl owering time and plant size on day 17. The predicted linear trend line any connection between a TF’s main effect on a trait and its and the statistical test results are shown. The average epistasis value for each single gene across all its epistatic pairs is plotted against that gene’s average epistatic effect on the trait. This allows us to investigate average single gene effect on the value as calculated against the respec- if there is any internal influence of single gene effects on the tive wild type. (A) Comparison of average main gene and epistatic gene direction and value of their epistatic interactions. If there is a effects on SC GLS (r = 20.610, P = 0.004). This relationship was signif- significant correlation, it would suggest that a gene’smainef- icant in both the presence and absence of the outlier point for SC GLS. (B) fl fects can provide information about epistatic networks. To do Comparison of average main gene and epistatic gene effects on ower- ’ ing time (r = 20.732, P , 0.001). (C) Comparison of average main gene this, we calculated each TF s average epistasis value for each and epistatic gene effects on flowering time on growth (r = 20.841, P , trait across all the pairwise combinations involving that TF. 0.001). Next, we correlated the TF’s estimated single mutant effects

Growth and Defense Coordination by TFs 537 Figure 10 Proposed model for TF coordina- tion of plant development and aliphatic GLS- based plant defense. External environmental and internal developmental signals are coor- dinately perceived via a group of TFs. Direct and indirect connections between these TFs allow for a coordinated response to these complex signals. A theoretically cohesive re- sponse is then transmitted to growth and defense via separate outputs from this network.

to their average epistasis value for each phenotype and chamber propose that the internal and external signals are coordi- combinations (Figure 9 and Supplemental Data Set 6). The nated and perceived via connections between these diverse analysis showed that within all three tested traits, there are TFs. These connections create a decision matrix whereby consistent negative correlations between a ’smain growth and aliphatic GLS are interpreted and coordinated. and epistatic effects. One possible interpretation of this result Independent connections to growth and aliphatic GLS then is that the genetic background is constraining the results that proceed from this matrix to maximize plant fitness in a given we have obtained. There may be a maximal effect size within a environmental setting (Figure 10). Further, this suggests that background that would cause larger main effect TFs to invari- TF networks provide an unappreciated potential to fine-tune antlyhavesmallerepistaticinteractions.Theprevalenceofneg- both growth and defense to optimize modern agriculture. ative epistasis in the SC GLS phenotype agrees with this Growth and defense vs. coordination possibility. Future work is required to understand if this obser- vation is a general property of this network or is a function of The canonical model for plant growth and plant defense is a the specific Col-0 accession in which it was conducted. trade-off model, which treats plant growth and defense as two competing biological processes under the assumption that the acquisition of elements and energy are limiting. As an alter- Discussion native hypothesis, growth and defense are two separable In this study, we tested 20 TF mutants and 48 paired double outputs of the plant’s regulatory system that must be coordi- mutants that influence aliphatic GLS accumulation to system- nated depending on the specific environment. The analysis of ically explore whether these TFs also influence plant growth the phenotypes in this TF collection and epistatic network in and flowering time. Seventeen of these 20 TFs significantly aliphatic GLS pathway supported the existence of the coor- influence plant growth and flowering time. Interestingly, the dination model. The complex interactions between growth key aliphatic GLS regulators MYB28 and MYB29 have little and defense depended on the specific perturbed TF, the en- influence on plant growth and flowering. In addition, no sim- vironment and their epistatic effects. We found no consistent ple mechanistic connection between the TFs’ effects on the evidence of any negative relationship between the accumu- accumulation of aliphatic SC GLS, growth, or flowering was lation of the SC GLS defense metabolites and any measure- found. These findings fit with the emerging coordination ment of growth. Future experiments will need to incorporate model whereby a network is dynamic and highly responsive more complex regulatory relationships to truly understand to the specific requirements of a specific environment. We how growth and defense are related in the field.

538 B. Li et al. One complication to the interpretation of these results is (grant DNRF99 to D.J.K.). S.M.B. was partially funded by a associated with the use of T-DNA mutants. Recent Howard Hughes Medical Institute faculty scholar fellowship. work shows that individual Arabidopsis T-DNA mutants can have chromosomal rearrangements that could potentially af- Literature Cited fect the phenotype via off-target effects linked to growth plei- otropy (Jupe et al. 2019). However, those are likely limited to Agrawal, A. A., 2011a Current trends in the evolutionary ecology of individual TFs and are unlikely to affect a whole collection of plant defence. Funct. Ecol. 25: 420–432. https://doi.org/10.1111/ 20 TFs. We show that the whole collection supported a lack of a j.1365-2435.2010.01796.x Agrawal, A. A., 2011b New synthesis-trade-offs in chemical ecology. correlation between growth and defense. Further, recent evi- J. Chem. Ecol. 37: 230–231. https://doi.org/10.1007/s10886-011- dence suggests that TFs that directly control GLS genes or GLS 9930-7 metabolites themselves can directly influence growth (Zhao Almeida-Cortez, J. S., B. Shipley, and J. T. Arnason, 1999 Do et al. 2008; Kerwin et al. 2011; Khokon et al. 2011; Wang plant species with high relative growth rates have poorer chem- – et al. 2011; Urbancsok et al. 2017; Zhu and Assmann 2017; ical defences? Funct. Ecol. 13: 819 827. https://doi.org/10.1046/ j.1365-2435.1999.00383.x Salehin et al. 2019). For example, MYB29 can control growth in Alonso,J.M.,A.N.Stepanova,T.J.Leisse,C.J.Kim,H.M.Chenet al., response to differential nitrate potentially by its direct interac- 2003 Genome-wide insertional mutagenesis of Arabidopsis thali- tion with the CCA1 promoter (Gaudinier et al. 2018). However, ana. Science 301: 653–657 (erratum: Science 301: 1849). https:// to fully understand how the plant is coordinating growth and doi.org/10.1126/science.1086391 defense will require a complete catalog of all TF-TF interactions Bates, D., M. Machler, B. M. Bolker, and S. C. Walker, 2015 Fitting linear mixed-effects models Using lme4. J. Stat. Softw. 67: 1–48. at the protein and promoter levels. This catalog would clarify if https://doi.org/10.18637/jss.v067.i01 genetic effects that we classify as indirect are in fact direct Bekaert, M., P. P. Edger, C. M. Hudson, J. C. Pires, and G. C. Con- interactions that, instead of being a single molecular step, i.e., ant, 2012 Metabolic and evolutionary costs of herbivory de- TF to promoter, utilize a longer cascade of connected molecular fense: systems biology of glucosinolate synthesis. New Phytol. – processes than are typically studied. 196: 596 605. https://doi.org/10.1111/j.1469-8137.2012.04302.x Beran, F., Y. Pauchet, G. Kunert, M. Reichelt, N. Wielsch et al., Using transcription factors to tune and optimize growth 2014 Phyllotreta striolata flea beetles use host plant defense and defense compounds to create their own glucosinolate-myrosinase system. Proc. Natl. Acad. Sci. USA 111: 7349–7354. https://doi.org/ Putative costs of plant defenses on growth have been inten- 10.1073/pnas.1321781111 sively researched, and recent findings suggest that it is pos- Bergelson, J., and C. B. Purrington, 1996 Surveying patterns in – sible to promote one without sacrificing the other (Campos the cost of resistance in plants. Am. Nat. 148: 536 558. https:// doi.org/10.1086/285938 et al. 2016; Wang et al. 2018). Our current study further Bidart-Bouzat, M. G., and D. J. Kliebenstein, 2008 Differential expands the future potential to optimize plant growth and levels of herbivory in the field associated with genotypic defense through TF manipulation. Specifically, we found that variation in glucosinolates in Arabidopsis thaliana. J. Chem. Ecol. a vast array of potential TFs can be identified and potentially 34: 1026–1037. https://doi.org/10.1007/s10886-008-9498-z Burow, M., B. A. Halkier, and D. J. Kliebenstein, 2010 Regulatory used to manipulate plant growth and defense. In our system, fi fi networks of glucosinolates shape Arabidopsis thaliana tness. mutations in more than half of the tested TFs could signi - Curr. Opin. Plant Biol. 13: 348–353. https://doi.org/10.1016/ cantly promote plant growth and defense together across di- j.pbi.2010.02.002 verse environmental conditions. Critically, this ability to Campos, M. L., Y. Yoshida, I. T. Major, D. de Oliveira Ferreira, S. M. promote both defense and growth is conditional on specific Weraduwage et al., 2016 Rewiring of jasmonate and phyto- environments. If this conditionality is true in other genes and chrome B signalling uncouples plant growth-defense tradeoffs. Nat. Commun. 7: 12570. https://doi.org/10.1038/ncomms12570 TFs involved in plant growth and defense, it will raise the Chan,E.K.F.,H.C.Rowe,andD.J.Kliebenstein,2010 Understanding importance of studying plants’ response to genetic manipu- the of defense metabolites in Arabidopsis thaliana using lation under different growth conditions. This is particularly genome-wide association mapping. Genetics 185: 991–1007. relevant to crop-breeding efforts in response to climate https://doi.org/10.1534/genetics.109.108522 change and environmental stresses. However, studies involv- Chan, E. K., H. C. Rowe, J. A. Corwin, B. Joseph, and D. J. Klieben- stein, 2011 Combining genome-wide association mapping and ing larger collections of TFs and epistatic interactions across transcriptional networks to identify novel genes controlling glu- even more diverse environments are needed to further un- cosinolates in Arabidopsis thaliana. PLoS Biol. 9: e1001125. derstand the coordination between growth and defense https://doi.org/10.1371/journal.pbio.1001125 across fluctuating environments. deLucas,M.,L.Pu,G.Turco,A.Gaudinier,A.K.Moraoet al., 2016 Transcriptional regulation of Arabidopsis polycomb repres- sive complex 2 coordinates cell-type proliferation and differentiation. Acknowledgments Plant Cell 28: 2616–2631. https://doi.org/10.1105/tpc.15.00744 Falk, K. L., J. Kastner, N. Bodenhausen, K. Schramm, C. Paetz et al., Funding for this work was provided by the National Science 2014 The role of glucosinolates and the jasmonic acid pathway in Foundation (awards IOS 1655810 and 1547796 to D.J.K.), resistance of Arabidopsis thaliana against molluscan herbivores. Mol. Ecol. 23: 1188–1203. https://doi.org/10.1111/mec.12610 the US Department of Agriculture National Institute of Food Fan, M., M. Y. Bai, J. G. Kim, T. Wang, E. Oh et al., 2014 The and Agriculture (Hatch project number CA-D-PLS-7033-H to bHLH transcription factor HBI1 mediates the trade-off between D.J.K.) and by the Danish National Research Foundation growth and pathogen-associated molecular pattern-triggered

Growth and Defense Coordination by TFs 539 immunity in Arabidopsis. Plant Cell 26: 828–841. https://doi.org/ leaf-chewing herbivore and an associated parasitoid. Phytochemistry 10.1105/tpc.113.121111 77: 162–170. https://doi.org/10.1016/j.phytochem.2012.01.005 Gaudinier, A., J. Rodriguez-Medina, L. Zhang, A. Olson, C. Liseron- Lambrix, V., M. Reichelt, T. Mitchell-Olds, D. J. Kliebenstein, and J. Monfils et al., 2018 Transcriptional regulation of nitrogen-associated Gershenzon, 2001 The Arabidopsis epithiospecifier protein metabolism and growth. Nature 563: 259–264. https://doi.org/ promotes the hydrolysis of glucosinolates to nitriles and influ- 10.1038/s41586-018-0656-3 ences Trichoplusia ni herbivory. Plant Cell 13: 2793–2807. Guo, R., W. Shen, H. Qian, M. Zhang, L. Liu et al., 2013 Jasmonic https://doi.org/10.1105/tpc.010261 acid and glucose synergistically modulate the accumulation of Li, B., A. Gaudinier, M. Tang, M. Taylor-Teeples, N. T. Nham et al., glucosinolates in Arabidopsis thaliana. J. Exp. Bot. 64: 5707– 2014 Promoter-based integration in plant defense regulation. Plant 5719. https://doi.org/10.1093/jxb/ert348 Physiol. 166: 1803–1820. https://doi.org/10.1104/pp.114.248716 Hemm, M. R., M. O. Ruegger, and C. Chapple, 2003 The Arabi- Li,B.,M.Tang,A.Nelson,H.Caligagan,X.Zhouet al., 2018 Network- dopsis ref2 mutant is defective in the gene encoding CYP83A1 guided discovery of extensive epistasis between transcription fac- and shows both phenylpropanoid and glucosinolate phenotypes. tors involved in aliphatic glucosinolate biosynthesis. Plant Cell 30: Plant Cell 15: 179–194. https://doi.org/10.1105/tpc.006544 178–195. https://doi.org/10.1105/tpc.17.00805 Huot, B., J. Yao, B. L. Montgomery, and S. Y. He, 2014 Growth- Lozano-Duran,R.,A.P.Macho,F.Boutrot,C.Segonzac,I.E.Somssich defense tradeoffs in plants: a balancing act to optimize fitness. et al., 2013 The transcriptional regulator BZR1 mediates trade-off Mol. Plant 7: 1267–1287. https://doi.org/10.1093/mp/ssu049 between plant innate immunity and growth. eLife 2: e00983. Jiao, Y., Y. Wang, D. Xue, J. Wang, M. Yan et al., 2010 Regulation https://doi.org/10.7554/eLife.00983 of OsSPL14 by OsmiR156 defines ideal plant architecture in Mikkelsen, M. D., B. L. Petersen, E. Glawischnig, A. B. Jensen, E. rice. Nat. Genet. 42: 541–544. https://doi.org/10.1038/ng.591 Andreasson et al., 2003 Modulation of CYP79 genes and glu- Joseph, B., J. A. Corwin, T. Zuest, B. Li, M. Iravani et al., cosinolate profiles in Arabidopsis by defense signaling path- 2013 Hierarchical nuclear and cytoplasmic genetic architec- ways. Plant Physiol. 131: 298–308. https://doi.org/10.1104/ tures for plant growth and defense within Arabidopsis. Plant Cell pp.011015 25: 1929–1945. https://doi.org/10.1105/tpc.113.112615 Pajerowska-Mukhtar, K. M., W. Wang, Y. Tada, N. Oka, C. L. Tucker Jupe, F., A. C. Rivkin, T. P. Michael, M. Zander, S. T. Motley et al., et al., 2012 The HSF-like transcription factor TBF1 is a major 2019 The complex architecture and epigenomic impact of molecular switch for plant growth-to-defense transition. Curr. plant T-DNA insertions. PLoS Genet. 15: e1007819. https:// Biol. 22: 103–112. https://doi.org/10.1016/j.cub.2011.12.015 doi.org/10.1371/journal.pgen.1007819 Pangesti, N., M. Reichelt, J. E. van de Mortel, E. Kapsomenou, J. Karban, R., and I. T. Baldwin, 1997 Induced Responses to Herbiv- Gershenzon et al., 2016 Jasmonic acid and ethylene signaling ory, University of Chicago Press, Chicago, IL. https://doi.org/ pathways regulate glucosinolate levels in plants during rhizo- 10.7208/chicago/9780226424972.001.0001 -induced systemic resistance against a leaf-chewing her- Kerwin,R.E.,J.M.Jiménez-Gómez,D.Fulop,S.L.Harmer,J.N.Maloof bivore. J. Chem. Ecol. 42: 1212–1225. https://doi.org/10.1007/ et al., 2011 Network quantitative trait loci mapping of circadian s10886-016-0787-7 clock outputs identifies -to-clock linkages in Arabi- Paul-Victor, C., T. Züst, M. Rees, D. J. Kliebenstein, and L. A. Turn- dopsis. Plant Cell 23: 471–485. https://doi.org/10.1105/tpc.110.082065 bull, 2010 A new method for measuring relative growth rate Kerwin, R., J. Feusier, J. Corwin, M. Rubin, C. Lin et al., can uncover the costs of defensive compounds in Arabidopsis 2015 Natural genetic variation in Arabidopsis thaliana defense thaliana. New Phytol. 187: 1102–1111. https://doi.org/10.1111/ metabolism genes modulates field fitness. eLife 4: e05604. j.1469-8137.2010.03325.x https://doi.org/10.7554/eLife.05604 R Development Core Team, 2014 R: A Language and Environment Kerwin,R.E.,J.Feusier,A.Muok,C.Lin,B.Larsonet al., 2017 Epistasis for Statistical Computing, R Foundation for Statistical Comput- x environment interactions among Arabidopsis thaliana glucosinolate ing, Vienna, Austria. genes impact complex traits and fitness in the field. New Phytol. Salehin, M., B. Li, M. Tang, E. Katz, L. Song et al., 2019 Auxin- 215: 1249–1263. https://doi.org/10.1111/nph.14646 sensitive Aux/IAA proteins mediate drought tolerance in Arabi- Khokon, M. A., M. S. Jahan, T. Rahman, M. A. Hossain, D. Muroyama dopsis by regulating glucosinolate levels. Nat. Commun. 10: 4021. et al., 2011 Allyl isothiocyanate (AITC) induces stomatal closure Schlaeppi, K., N. Bodenhausen, A. Buchala, F. Mauch, and P. Rey- in Arabidopsis. Plant Cell Environ. 34: 1900–1906. https:// mond, 2008 The glutathione-deficient mutant pad2–1accumu- doi.org/10.1111/j.1365-3040.2011.02385.x lates lower amounts of glucosinolates and is more susceptible to Kliebenstein, D. J., 2016 False idolatry of the mythical growth the insect herbivore Spodoptera littoralis. Plant J. 55: 774–786. vs. immunity tradeoff in molecular systems plant pathology. https://doi.org/10.1111/j.1365-313X.2008.03545.x Physiol. Mol. Plant Pathol. 95: 55–59. https://doi.org/10.1016/ Searle, S. R., F. M. Speed, and G. A. Milliken, 1980 Population j.pmpp.2016.02.004 marginal means in the linear model: an alternative to least Kliebenstein, D. J., J. Kroymann, P. Brown, A. Figuth, D. Pedersen squares means. Am. Stat. 34: 216–221. et al., 2001 Genetic control of natural variation in Arabidopsis Segrè, D., A. Deluna, G. M. Church, and R. Kishony, 2005 Modular thaliana glucosinolate accumulation. Plant Physiol. 126: 811– epistasis in yeast metabolism. Nat. Genet. 37: 77–83. https:// 825. https://doi.org/10.1104/pp.126.2.811 doi.org/10.1038/ng1489 Kliebenstein, D., D. Pedersen, B. Barker, and T. Mitchell-Olds, Shannon, P., A. Markiel, O. Ozier, N. S. Baliga, J. T. Wang et al., 2002a Comparative analysis of quantitative trait loci controlling 2003 Cytoscape: a software environment for integrated models glucosinolates, myrosinase and insect resistance in Arabidopsis of biomolecular interaction networks. Genome Res. 13: 2498– thaliana. Genetics 161: 325–332. 2504. https://doi.org/10.1101/gr.1239303 Kliebenstein, D. J., A. Figuth, and T. Mitchell-Olds, 2002b Genetic Singh,K.,R.C.Foley,andL.Onate-Sanchez, 2002 Transcription architecture of plastic methyl jasmonate responses in Arabidop- factors in plant defense and stress responses. Curr. Opin. Plant Biol. sis thaliana. Genetics 161: 1685–1696. 5: 430–436. https://doi.org/10.1016/S1369-5266(02)00289-3 Koricheva, J., 2002 Meta-analysis of sources of variation in fitness costs Sønderby, I. E., B. G. Hansen, N. Bjarnholt, C. Ticconi, B. A. Halkier of plant antiherbivore defenses. Ecology 83: 176–190. https:// et al., 2007 A systems biology approach identifies a R2R3 MYB doi.org/10.1890/0012-9658(2002)083[0176:MAOSOV]2.0.CO;2 gene subfamily with distinct and overlapping functions in reg- Kos, M., B. Houshyani, R. Wietsma, P. Kabouw, L. E. Vet et al., ulation of aliphatic glucosinolates. PLoS One 2: e1322. https:// 2012 Effects of glucosinolates on a generalist and specialist doi.org/10.1371/journal.pone.0001322

540 B. Li et al. Sønderby,I.E.,M.Burow,H.C.Rowe,D.J.Kliebenstein,andB. glucosinolate activation. Plant Physiol. 147: 415–428. https:// A. Halkier, 2010a A complex interplay of three R2R3 MYB doi.org/10.1104/pp.107.115279 transcription factors determines the profile of aliphatic glu- Zalucki, M. P., J. M. Zalucki, L. E. Perkins, K. Schramm, D. G. cosinolates in Arabidopsis. Plant Physiol. 153: 348–363. Vassao et al., 2017 A generalist herbivore copes with special- https://doi.org/10.1104/pp.109.149286 ized plant defence: the effects of induction and feeding by hel- Sønderby, I. E., F. Geu-Flores, and B. A. Halkier, 2010b Biosynthesis icoverpa armigera (Lepidoptera: noctuidae) larvae on intact of glucosinolates–gene discovery and beyond. Trends Plant Sci. Arabidopsis thaliana (Brassicales) plants. J. Chem. Ecol. 43: 15: 283–290. https://doi.org/10.1016/j.tplants.2010.02.005 608–616. https://doi.org/10.1007/s10886-017-0855-7 Sussman, M. R., R. M. Amasino, J. C. Young, P. J. Krysan, and S. Zhao, Z., W. Zhang, B. A. Stanley, and S. M. Assmann, 2008 Func- Austin-Phillips, 2000 The Arabidopsis knockout facility at the tional proteomics of Arabidopsis thaliana guard cells uncovers University of Wisconsin-Madison. Plant Physiol. 124: 1465– new stomatal signaling pathways. Plant Cell 20: 3210–3226. 1467. https://doi.org/10.1104/pp.124.4.1465 https://doi.org/10.1105/tpc.108.063263 Urbancsok, J., A. M. Bones, and R. Kissen, 2017 Glucosinolate- Zhu, M., and S. M. Assmann, 2017 Metabolic signatures in re- derived isothiocyanates inhibit Arabidopsis growth and the po- sponse to abscisic acid (ABA) treatment in Brassica napus guard tency depends on their side chain structure. Int. J. Mol. Sci. 18: cells revealed by metabolomics. Sci. Rep. 7: 12875. https:// 2372. https://doi.org/10.3390/ijms18112372 doi.org/10.1038/s41598-017-13166-w Wang,R.S.,S.Pandey,S.Li,T.E.Gookin,Z.Zhaoet al., Züst, T., and A. A. Agrawal, 2017 Trade-offs between plant growth and defense against insect herbivory: an emerging 2011 Common and unique elements of the ABA-regulated tran- – scriptome of Arabidopsis guard cells. BMC 12: 216. mechanistic synthesis. Annu. Rev. Plant Biol. 68: 513 534. https://doi.org/10.1186/1471-2164-12-216 https://doi.org/10.1146/annurev-arplant-042916-040856 Züst, T., B. Joseph, K. K. Shimizu, D. J. Kliebenstein, and L. A. Turn- Wang,J.,L.Zhou,H.Shi,M.Chern,H.Yuet al., 2018 A single bull, 2011 Using knockout mutants to reveal the growth costs of transcription factor promotes bothyieldandimmunityinrice.Sci- defensive traits. Proc. Biol. Sci. 278: 2598–2603. https://doi.org/ ence 361: 1026–1028. https://doi.org/10.1126/science.aat7675 10.1098/rspb.2010.2475 Wentzell, A. M., and D. J. Kliebenstein, 2008 Genotype, age, tissue, and environment regulate the structural outcome of Communicating editor: J. Birchler

Growth and Defense Coordination by TFs 541